Above the Clouds: From Grids to Service-
oriented Operating Systems
Lutz SCHUBERTa,1, Alexander KIPPa and Stefan WESNERa
aHLRS – University of Stuttgart, Germany
Abstract. Over recent years, resource provisioning over the Internet has moved
from Grid to Cloud computing. Whilst the capabilities and the ease of use have
increased, uptake is still comparatively slow, in particular in the commercial
context. This paper discusses a novel resource provisioning concept called
Service-oriented Operating Systems and how it differs from existing approaches of
Grids and Clouds. The proposed approach aims for making applications and
computers more independent of the underlying hardware and increase mobility and
performance. The base architecture and functionality will be detailed in this paper,
as well as how such operating systems could be deployed in future workspaces.
Keywords. operating systems, network, grid, service oriented architecture, many
core, mobile grids, distributed computing
Introduction
The Grid concept is understood in this paper as a set of distributed resources integrated
in a fashion allowing remote execution of processes and applications with different
requirements towards the underlying resources, such as computational power. Next to
execution of tasks, it is also often considered a space for storing vast data amounts.
These types are often referred to as computational or storage Grids. Depending on
whether execution / storage is supposed to be stable non-regarding changes in the Grid
structure (i.e. failure of individual resource nodes), we speak of managed grids.
Though these do not comprise the full usage scope of Grids, it nonetheless reflects
the base commonality of most use cases, namely the capability to provide (managed)
resources in a fashion that they can be used remotely with little to no impact on the
local execution of processes, respectively extending the capabilities of local resources.
Cloud computing has recently emerged from this movement as a means to provide
in particular computational resources, even though storage clouds are gaining in
popularity too. From the Grid perspective, Clouds are only passive resources in the
sense of nodes in the computational or data grid case – they do not offer the enhanced
capabilities of Virtual Organisations or similar, nor do they easily plug in into existing
grid structures (such as GT4). Typically cloud and similar resource providers expose
their own proprietary APIs which imply that the resources are used in a more manual
fashion than originally envisaged by the Grid. From this point of view, clouds are only
extended single Grid nodes. However they can be seen as an intermediary step to
enabling dynamic outsourcing in a form that extends local resources.

1
Corresponding Author: Lutz Schubert, HLRS – High Performance Computing Centre, University of
Stuttgart. Nobelstr. 19, 70569 Stuttgart, eMail: schubert@hlrs.de

This paper will examine the current problems in resource provisioning and uptake
in commerce, as well as proposing a means to overcome these problems (section 1 & 2).
We will present the notion of Service Oriented Operating Systems (SOS) that integrate
dynamic resource pools and execute processes across these (section 3 & 4). We will
then compare this model with current related approaches (section 5) and conclude by
examining the relevance of resource fabrics for the future internet.
1. From Grids to Clouds and their limitations
Early Grid solutions had been introduced to overcome the limitations of computing
systems and integrate geographically dispersed resources into a metacomputer [19].
This concept had been further extended to allow for integration of more heterogeneous
resources. Prominent realisations of this are the EGEE [20] and DEISA [21] infra-
structures. While they are well perceived and used within the scientific community
their uptake for business or general purpose applications remains quite limited.
The move from proprietary solutions towards Service Oriented Grids [22] [23] has
increased uptake within the community as critical aspects such as commercial level
security, exploitation of Web Service standards and frameworks decreased the entry
level border for programmers and companies significantly and is further developed
utilizing semantic technology towards Service Oriented Knowledge Utilities [24].
Grids realising cross-organisational collaborations remain still comparatively complex
and competing standards for key aspects make investment in this technology risky.
Cloud computing addresses the major limitations of Grid solutions and convinces
by its (seeming) simplicity. They pick up the concept of Virtual Hosting Environments
[23] [2] and realise them in an efficient and easy to use fashion. In order to be
commercially viable, outsourcing approaches like Clouds have to take extreme security
and legal precautions, as typically the data transported to and computed on external
resources is sensitive to the respective customer. The complexity of outsourced compu-
tational resources was one of the reasons to make the approach of grid- / web service
based Virtual Organisations [14] so attractive for enterprises.
Existing Cloud solutions only partially address such business requirements so far
and clearly attract users without specific requirements e.g. related to the confidentiality
of their data. The realization of Grids over Clouds (cf. Figure 3) would enable such
combined features but require additional mechanisms to bridge the gap between the
Grid and the Cloud environment.
Current approaches e.g. by Amazon [5] or SalesForce’ [4] suffer in particular from
their proprietary APIs and their clear disjunction from local machines: whilst network
storage can already be easily integrated into the user’s operating system given that it is
static, resources, services and applications on a network level cannot. Even though the
web service approach would allow for this integration due to its standard interface
approach, usage is not easy and typically comes at the cost of speed.
Web Services and Grid put the user in control of the actions and integrations –
something that very few users are willing to put up with, let alone have the knowledge
and capabilities to do so. For low level resources this is furthermore not even required.
Service-oriented Operating Systems treat resources as what they are: resources, i.e.
they are treated by the OS in the same way as local devices. As we will show in the
discussion section, even web based applications effectively similar to computational
resources with very limited capabilities (namely execution of the respective process).